How Do We Quantify the Value of AI?

Yesterday I facilitated a short workshop around a big open question:

How will we quantify the value of AI?

It sparked some interesting debate, but one follow-up question stood out to me:

How did we quantify value before AI, and do we really need to reinvent the wheel?

We talked about the tangible things we already use to understand value: cost reduction, productivity, time to market, revenue, customer satisfaction and quality, but there is an important challenge.

Productivity doesn't necessarily equal value.

AI might let us produce code faster, but if that code creates more defects, increases technical debt, introduces security concerns, or becomes difficult to maintain, have we really created value?

Perhaps that's why we need to look beyond the immediate productivity gain and consider the longer-term impact. Something that looks like a huge efficiency win today might look very different three or six months from now.

A baseline is important too. We need to understand what happened before AI and compare it with what happens after. Otherwise, how do we know whether AI actually made a difference?

We also recognised that AI value won't be consistent. Different models, tools, teams and use cases will produce different results. And not everything is easily quantifiable. Time and cost are relatively straightforward to measure; knowledge gained, learning and the wider impact on people and how we work are much harder to put a number against.

So perhaps the answer isn't to create an entirely new framework for measuring AI.

Maybe we simply need to apply the principles we already use to measure value and ask a slightly different question:

What problem are we trying to solve, what does success look like, and has AI helped us achieve it?

AI may be changing how we work, but it may not need to change how we think about value.

---

09/09/2026 Workshop Output:

Here are the main outputs from our session.

The main themes were:

• Start with the problem, not the AI. Link value to the problem or outcome the AI is helping solve.

• Measure tangible business impact, including:

  • Cost reduction

  • Revenue

  • Productivity

  • Time to live / delivery

  • Customer satisfaction (CSAT)

  • Quality and defect reduction

• Quality needs to go beyond code output. Measures could include fewer defects, maintainability, security and the longer-term value of AI-generated code.

• AI value varies by model and use case. Improvements shouldn’t be assumed to be consistent across different AI models or applications.

• Use a baseline. Compare AI-enabled performance with the traditional approach rather than measuring it in isolation.

• Consider the longer term. For example, assess the maintainability of AI-generated code after three months, rather than only measuring the initial productivity gain.

• Automation of measurement is important. Wherever possible, capture metrics automatically rather than relying on subjective/manual assessment.

• Not everything is easily quantifiable. The group distinguished between:

  • Hard metrics: time and cost

  • Softer metrics: knowledge gained and social impact

The emerging principle

AI value shouldn’t simply be measured by how much code is produced or how quickly it is produced.

Instead, measure whether AI creates better outcomes, at lower cost, with improved quality and sustainable long-term value, compared with what would have happened without AI.

And importantly, “Have we made the right thing easier, or have we simply made the wrong thing faster?”